Supporting material
Desk-researched interview 2: Faster coding, slower handoffs
Desk research, not original interviews. These question sets synthesize published accounts; no practitioner was interviewed for this book.
Question: Which work became faster with AI, and which downstream team received more work as a result?
Method: Desk research from published engineering accounts and survey commentary; no firsthand interviews were conducted for this book. Research snapshot: November 2026.
What the published accounts say
- Salesforce's Hyperforce Runtime Platform (HRP) engineering account reports faster local development: a 32% improvement in inner-loop efficiency in its initial three months, alongside a 26% gain in team throughput. Across roughly 500 engineers, more code entered the delivery path; PR review cycles lengthened, and production delivery depended on approvals from other teams. The article says the teams shifted common anti-pattern checks earlier into local development, introduced review tooling and worked with approval-gating partner teams. It does not quantify extra hours worked by each reviewer or name a single specific downstream department. Its before/after period also included change-approval optimizations, so the throughput change cannot be credited entirely to AI. Salesforce Engineering, 10 November 2026.
- Atlassian's 2026 developer-experience commentary describes an unnamed engineering leader whose team received an increased workload from upstream teams producing more with AI. Its survey summary says 87% of developers reported shipping faster with AI while 72% reported the same or more friction. That story identifies a downstream team generically, not its name, workload in hours, or measured customer consequence. Atlassian, 2026.
Synthesis, not testimony
In the named Salesforce case, coding and PR preparation became faster; human reviewers and approval-gating partner teams faced more incoming work. Salesforce's account directly says review cycles became longer; an increase in total reviewer hours per team is plausible but not measured there. It would be misleading to claim a measured transfer to QA or security specifically. The primary outcome at risk is efficient end-to-end flow, with reliable review quality and engaging reviewer workload as guardrails.
An investigation should count arrivals and service capacity at each stage, not just commits: PRs arriving per week, review queue age, reviewer time, requests to security/QA/operations where applicable, approval wait, deployment lead time and defects. Segment by change size and risk. Ask whether early AI review reduces human effort or merely adds another review step.
Questions to ask a practitioner
- Which task became faster (design, coding, tests or documentation), compared with what baseline?
- Which named team or rotation received the additional PRs, approvals or defects? How did its queue and hours change?
- Did customer lead time improve after all required review and release gates, without increasing escaped defects or on-call load?
Follow-up desk-researched responses
1. Which task and baseline? Salesforce HRP reports 32% better inner-loop efficiency during the first three months of its AI adoption program and a fall in cycle time from 22 to 11 days across comparison periods. These are team-level before/after delivery measures, not a controlled time trial for a specific coding, design or testing task; approval-process changes occurred in the same period. A separate Salesforce engineering account says AI-assisted work reduced time-to-code while code volume rose about 30% and PR latency rose. HRP account; review-system account.
2. Which named downstream team or rotation, and how many hours? Salesforce names senior human PR reviewers as the receiving role and mentions approval-gating partner teams, but does not identify a particular rotation or publish a queue-age series or hours per reviewer. Its review-system account reports quarter-over-quarter growth in review latency, PRs exceeding 20 files and 1,000 lines, and frequent context switching by senior reviewers. Those observations demonstrate the constraint; they do not quantify added labor or prove that all additional code was AI-authored. Salesforce review-system account.
3. Did value reach customers safely? Salesforce reports improved cycle time and a steady maintainability score in HRP, but not matched customer release-to-use time, escaped defects and on-call burden for the same changes. A separate Spotify account reports more than double the merged PRs year over year and warns that verification controls struggled to keep up; it says its incident review found no material direct AI-authored contributor among incidents reviewed so far. Neither account proves a controlled improvement in customer value without hidden downstream costs. HRP account; Spotify quality account.
Evidence gap: The cited accounts do not provide a single organization-wide ledger connecting AI minutes saved upstream to hours added downstream; Salesforce offers a direct bottleneck account, not a controlled causal estimate.